Hi Chuck, These patches build ontop of your recent series, here: https://lore.kernel.org/linux-nfs/20250909190525.7214-1-cel@xxxxxxxxxx/ The first sunrpc patch is carry-over from earlier iterations of my NFSD DIRECT patchsets. I think it needed to be sure we have adequate pages to handle expanding misaligned READ to fit in max payload. But maybe not? The 2nd "NFSD: Implement NFSD_IO_DIRECT for NFS WRITE" patch is a refactored version of my earlier NFS WRITE support. I tried to honor the types of changes I saw in your rewrite of the NFSD_IO_DIRECT NFS READ support. So I avoided needless maths and variables that were only for the benefit of excessive tracepoint complexity (which I also removed), while also establishing cleaner code that you can iterate on further if you think something else needed. LOC isn't much smaller but maybe you or others will see something that can be elided -- the good news is this NFSD DIRECT WRITE code stands on its own like the NFSD DIRECT READ does. Not sure what NFSD DIRECT WRITE gray areas you think we need to work through further in NFSD (but I could imagine there being room for common Linux vfs code or other FS-specific code improvements to make buffered vs direct safer). I know others have repeated concern but in practice I haven't had any issues with this NFSD DIRECT WRITE suppoort. So I welcome further discussion on what else needed. Thanks, Mike Mike Snitzer (2): sunrpc: add an extra reserve page to svc_serv_maxpages() NFSD: Implement NFSD_IO_DIRECT for NFS WRITE fs/nfsd/debugfs.c | 1 + fs/nfsd/trace.h | 1 + fs/nfsd/vfs.c | 215 ++++++++++++++++++++++++++++++++++++- include/linux/sunrpc/svc.h | 5 +- 4 files changed, 216 insertions(+), 6 deletions(-) -- 2.44.0
